Background

Diffuse large B-cell lymphoma (DLBCL) is the most prevalent but clinically and genetically heterogeneous group of non-Hodgkin lymphomas (NHL). LymphGen classification (Wright, George W., et al., 2020) is a widely used molecular subtyping algorithm in academic field, capable of categorizing DLBCL into seven genetic subtypes by utilizing next-generation sequencing (NGS) data, which include MCD, BN2, N1, EZB (MYC+/-), ST2, A53, and Other (unclassifiable). Each subtype exhibits unique molecular spectrum and prognostic characteristic, and thus responds differently to distinct targeted therapies. However, the practical application of this algorithm in clinical scenario requires further exploration.

Materials and Methods

A total of 200 patients diagnosed with DLBCL were enrolled in this study. NGS was conducted on fresh or paraffin-embedded tissue samples to assess variants in 238 lymphoma-associated genes, including the core genes of the LymphGen algorithm. The classification was based on results of mutations, copy number variations (CNV), and translocations (MYC, BCL2 and BCL6 gene).

Results

Using online LymphGen classification algorithm, 61% of patients can be classified, compared with 63.1% reported from the literature, In summary, we have identified MCD (18.0% vs 13.2%), BN2 (15.0% vs 16.1%), EZB (12.5% vs 13.2%), ST2 (9.5% vs 4.7%), N1 (2.5% vs 2.8%), A53 (0% vs 6.6%), genetically composite (3.5% vs 5.7%), and Other (39.0% vs 36.9%) compared with the literature. While most of the subtypes displayed a comparable distribution to those reported, MCD subtype showed a slightly higher proportion and ST2 subtype showed a notably higher proportion in this study, whereas A53 subtype was absent.

Additionally, 21 out of 200 DLBCL patients were found to be transformed and subjected to further analysis. Among these cases, 23.8% of EZB, 14.3% of N1, 9.5% of ST2, 4.8% of BN2, 4.8% of MCD, and 42.9% of Other were identified. EZB and N1 subtypes were enriched in this group, which exhibited genetic similarities to FL and CLL, respectively.

Therefore, we speculated that the following factors may account for the discrepancies between our results and those reported:

1. The training and validation data for the model reportedwere based on whole-exome sequencing, compared withmulti-gene targeted sequencing that was utilized in this study. Differences in the scope and depth between the two methodology may lead to inconsistancy in both the amount as well as the type of alterarion detected, potentially explaining the absence of the A53 subtype in this study.

2. 20% of samples in this study lacked results ofBCL2, BCL6, and MYC rearrangements, which can affect the performanceof algorithm.

3. The LymphGen algorithm didnot account for variants of allele frequency (VAF), whereas 134 (67%) paraffin-embedded tissue sampleswere collected in this study. Thus some low VAFs may interfere with classification outcomes.

4. The dataset used in the literature and thisstudy werefrom different ethic groups.

5. The online version of the LymphGen algorithm we used has been updated in real-time, potentially leading to performance differences.

Conclusion

In summary, our results showed a significant higher proportion of ST2 subtype and an absence of A53 subtype compared to literature. Among 21 transformed DLBCL samples, EZB and N1 subtypes were more prevalent than others. Our findings suggest that while the LymphGen algorithm demonstrates an excellent performance in research settings, in real-world scenario endeavors to adjust to and accommodate specific patient population and clinical outcomes were certainly required. Furthermore, classification of transformed DLBCL samples indicated that LymphGen algorithm may have potential application values in understanding of disease transformation in DLBCL patients.

Disclosure

No relevant conflicts of interest to declare.

Disclosures

No relevant conflicts of interest to declare.

This content is only available as a PDF.
2024
Sign in via your Institution